cursor-editlisted

# Cursor Edit ## Overview This skill outsources **only the act of editing files** to Cursor's headless CLI (`cursor-agent`). Use the same contract from Claude Code, Codex, or any other AI agent. - **Calling AI agent (brain)**: reads and searches code, decides what to change, writes edit instructions, and verifies the result - **Cursor (hands)**: rewrites files according to the received instructions The goal is to move heavy editing work to Cursor, reducing token usage in the calling AI agent and speeding up the overall task. This only pays off for **larger edits**. For small changes, the cost of writing self-contained instructions and reviewing `git diff` can exceed the cost of direct editing. Do not force this skill onto trivial one-line or few-line edits. ## When To Use These tendencies apply: - **Not suitable for trivial edits (one to a few lines)**. Startup, instruction writing, and diff review often cost more than direct editing by the calling AI agent. - **Best for edits that are large, pattern-heavy, and have good sibling files to imitate**. Example: adding tests that follow an existing test suite. `cursor-agent` can closely follow good sibling examples and often produce convention-compliant results quickly. - **Be careful with new features or non-obvious assumptions when no good sibling example exists**. `cursor-agent` tends to pattern-match without semantic validation, so correctness is not guaranteed. In this case, the calling AI agent must separately verify